Circular import: four consolidation submodules cannot be imported first (headless_authoring hub)

[cdeust]

Reproduction

Each of these fails in a fresh interpreter (verified on main, 3a1ba1e, 2026-07-28):

$ python -c "import mcp_server.handlers.consolidation.candidate_scan"
ImportError: cannot import name '_collect_anchor_candidates' from partially
initialized module ... (most likely due to a circular import)

Same failure shape for claude_cli (_build_argv), cycle_orchestration
(run_headless_authoring_cycle), and drain_operations
(drain_all_gaps_on_page).

Cause

The four submodules each do from . import headless_authoring as _root at
module top while headless_authoring imports names back from them — the
same load-order-sensitive hub cycle as #233's synaptic pair, one package
over. Importing headless_authoring first resolves it, which is why the
suite stays green; any consumer (REPL, script, future module) that imports
a submodule directly fails. tests_py/handlers/test_s110_sweep_handlers.py
already has to import headless_authoring first with a comment noting the
load-order sensitivity.

Surfaced by the #197 family-4 (PLC0415) sweep's per-module import
verification, which imports every mcp_server module in a fresh
interpreter. The sweep touched these files (hoisting unrelated lazy
imports) but deliberately did not restructure the hub cycle
(behavior-preserving mandate; a seam-creating refactor is its own PR per
coding-standards §15.1). The sweep's baseline check proves these four
failures are pre-existing on main, byte-for-byte identical before and
after the hoist.

Acceptance criteria

1. python -c "import mcp_server.handlers.consolidation.<m>" succeeds in
   a fresh interpreter for all four modules (regression test iterating them
   in subprocesses, as Circular import: synaptic_plasticity_hebbian cannot be imported first #233 criterion 1).
2. The _root back-reference is replaced by a leaf module (or explicit
   parameter passing) so no submodule imports the hub at load time.
3. Existing suite passes unchanged.
4. The load-order comment/workaround in
   tests_py/handlers/test_s110_sweep_handlers.py becomes unnecessary and
   is removed.

Refs #233 (same defect class), #197 (surfaced by family 4).